LLM09: Misinformation · 2025

The model produces false or misleading information that appears convincing; overreliance turns these errors into harmful decisions or actions.

Summary

The model produces false or misleading information that appears convincing; overreliance turns these errors into harmful decisions or actions.

How the attack works

The model invents a convincing reference; a person accepts the answer without checking its source and makes a wrong decision.

Example

Vulnerable · Pseudocode

Prompt: What is our return deadline?
Answer: 90 days, per policy section 8.
# No such section exists

Fix: Ground answers in verifiable sources, check citations and require human review for important decisions.

Prevention

Ground responses in trusted sources.
Require cross-checking and human oversight.
Automatically validate critical results and generated code.

Lineage

2025 · LLM09: Misinformation Moves up to rank 7 in 2026
2026 · LLM07: Misinformation Moves up from rank 9

2025:LLM09 → 2026:LLM07: Misinformation moves up from rank 9 to 7.

Sources

Open official source ↗

Buy me a coffee